Images are 2-D, how to adapt this for 1-D data ?

[mayank64ce]

No description provided.

[lucidrains]

what kind of data are you working with?

[mayank64ce]

Its like (N, D) shaped data where say N = number of frames of a video and D is the number of parameters extracted from each frame. I am struggling to make the prediction temporally consistent and thats how I found this 3D Unet Idea interesting. I am not sure if UNet 3d will work for me.

[lucidrains]

@mayank64ce i see, but could you not just use a 1d unet? i have one here, by popular request some time ago https://github.com/lucidrains/denoising-diffusion-pytorch/blob/main/denoising_diffusion_pytorch/denoising_diffusion_pytorch_1d.py

[mayank64ce]

I am using unet_1d from diffusers as of now, I am unable to get the generated vectors to be temporally coherent.